linear mixed integer ppg grogramming for topology ... · pdf filelinear mixed integer ppg...
TRANSCRIPT
Linear mixed integer programming g p g gfor topology optimization of
trusses and plates
Makoto Ohsaki and Ryo Watada
Kyoto University, JAPAN
Michell trussMichell truss
A.G.M. Michell, The limits of economy inframe structures 1904
Infinite number of members in
frame structures, 1904.
W Prager A note on discretizedprincipal stress line
W. Prager, A note on discretized Michell structure,1974
Finite number of membersConsider nodal cost
Ground structure approach to topology optimization
W. Dorn et al., Automatic design of optimal structures, 1964
1 A i ll th d d b1. Assign all the nodes and members.
2. Optimize cross‐sectional areas as continuous variables.
3. Remove unnecessary members and nodes.y
Optimization under stress constraints
Single loading condition
Statically determinateMultiple loading condition Statically determinate fully stressed
Statically indeterminate ygenerally not fully stressed
Formulation of topologyFormulation of topology optimization under stress constraintsoptimization under stress constraints
Minimize total structural volume
UkL σσσ ≤≤ )(A 0≥Afs. t. iii σσσ ≤≤ )(A 0≥iAfor
Stress constraints are given only forStress constraints are given only for existing members
iA : cross-sectional area of member i
)(Akσ : stress of member i for load k)(Aiσ : stress of member i for load k
Multiple loading conditionsMultiple loading conditions
Optimal topology
Member 3 does not existMax. stress ratio = 4.03
V = 12.812
Fully stressed design Optimal topology under multiple loading conditions cannot beloading conditions cannot be obtained by conventionalground structure approach
V = 15.986
Difficulties in topology optimization of trusses
• No stress constraint for non‐existing member.
• Discontinuity in problem formulation.St t i t dd l di– Stress constraint suddenly disappear as area approach 0.
• Optimal solution may exist at a singular Feasible regiongpoint (cusp).
Optimal solution
Unstable optimal solutionUnstable optimal solution (member buckling)(member buckling)
i ioptimize
Fix node 2 different slenderness ratioFix node 2 different slenderness ratiodifferent stress bound
Infeasible optimal solutionsInfeasible optimal solutions
Intersection of members
Very thin member
Unstable node
d fMixed integer programming for truss topology optimizationtruss topology optimization
with discrete variableswith discrete variables
Topology optimization problemTopology optimization problem
minimize V a l= ∑ Structural volumeminimize
subject to
j jj
V a l=
=
∑Bs f
Structural volume
equilibrium
min max
subject to j j j j ja s a
Eσ σ≤ ≤
Bs fStress constraint
q
j Tj j j
j
Es a
l= b uForce constraint Nonlinear relation
0j
ja ≥V i bl
Select from list of available sectionsjaVariables: s, u, ja
Select from available listSelect from available list{0 }a a a∈ =A 1{0, , , }
jj j j jNa a a∈ =A
⎧0-1variable
1 if 0 otherwise
j jkjk
a ax
=⎧= ⎨⎩0 otherwise⎩
Express using jkxN N
1 1
, 1j jN N
j jk jk jk jk k
a x a x x= =
= = ≤∑ ∑
1,
jNj T
jk jk jk j j jkk
Es x a s s
l= =∑b u
1kjl =Stress-displ. relation
Linearization of stress disp relationLinearization of stress‐disp. relation
Linearization of nonlinear constraints(Stolpe and Svanberg)( p g)
0 1TiEs x a x orb umin max
ik ik ikx c s x c≤ ≤
, 0 1iik ik ik i ik
i
s x a x orl
= =b u
min T max(1 ) (1 )
ik ik ik
ik ik j ik iki
Ex c a s x cl
− ≤ − ≤ −b uil
Cmax, Cmin: sufficiently large/small value
Mixed integer linear programming (MILP)
Optimization of bridge truss from traditional configurations
SelectionTraditional configurations
Select from three types
Howe truss Platt truss
⇒ combination of traditionalconfigurations.Warren truss K‐truss
Optimization results by CPLEXOptimization results by CPLEXCase 1: 40.7 10 V=65.410 CPU = 67900(sec)f −= ×
4Case 2: 41.0 10 V=74.610 CPU = 14600(sec)f −= ×
Case 3: 41.4 10 V=81.203 CPU = 10600(sec)f −= ×
Warren truss for small loads.
⇒ Howe truss near support for large loads. ( {0.0,1.0,1.8})=A
CPU time strongly depends on load magnitude.
Optimization of space trussOptimization of space truss
Loaded node
Optimization of space truss• Ground structure Selection pattern
Traditional configurationsTraditional configurations
A: Schuwedler domeB: Lamella domeB: Lamella dome
Schuwedler dome Lamella dome
Optimization resultsOptimization resultsCase 1: 41.0 10 V=1404.3f −= × Case 2: 43.0 10 V=1441.1f −= ×
4 4Case 3: 41.0 10 V=1404.3f −= × Case 4: 410.0 10 V=1527.2f −= ×
Lamella dome for small loads.
( {0.0,1.0,2.0})=A
⇒ Schuwedler dome in perimeter region for large loads.